How geometric similarity analysis works for figurative trade marks

The problem with pixel comparison

Most approaches to comparing figurative trade marks treat the marks as images. Two images are compared pixel by pixel, or their pixel data is compressed into a feature vector and the vectors are compared. This works reasonably well for near-identical marks, but it fails in exactly the situations where trade mark conflicts actually arise: marks that share structural characteristics but differ in surface detail.

A circle with a line through it and an oval with a line through it may represent a genuine conflict in trade mark terms. A pixel comparison might rate them as only moderately similar, because the pixel distributions are different. A tribunal, looking at the same two marks, would likely find them visually similar, because the structural composition is the same.

From characters to marks

The geometric comparison approach used by Markscope was originally developed for a different problem: identifying visually confusable characters in the Unicode Standard. When two characters from different scripts look similar enough to deceive a reader, they represent a security risk in domain names, email addresses, and source code identifiers.

The published research on this problem introduced a method for measuring structural similarity between vector outlines. Rather than comparing pixels, the approach works directly with the mathematical curves that define a shape. The same principles that can determine whether a Cyrillic letter is confusable with a Latin letter can determine whether two figurative trade marks share enough structural similarity to create a likelihood of confusion.

How geometric comparison works

The approach treats each mark as a set of vector outlines and analyses the structural relationships within those outlines. Without describing the specific implementation in detail, the general principle is straightforward: instead of asking “do these marks look the same as images?”, the system asks “do these marks have the same structural composition?”

This means comparing the spatial relationships between the elements of each mark, the proportions, the angles, the relative positions of features, and the overall geometric arrangement. Two marks might use different line weights, different colours, or different levels of detail, but if their underlying geometric structure is equivalent, the system identifies them as structurally similar.

The comparison is deterministic. Given the same two marks, it will always produce the same result. There is no training data, no neural network weights that might shift between versions, and no probability distribution to interpret. The output is a structured explanation of what is geometrically similar between the two marks and to what degree.

Why this matters for trade mark analysis

The UK Intellectual Property Office and the courts assess visual similarity as one of the key factors in determining likelihood of confusion. The assessment is holistic, considering the marks as a whole while also noting distinctive and dominant elements. This is fundamentally a structural assessment, not a pixel-level one.

A geometric comparison approach aligns more closely with how human assessors and tribunals actually evaluate figurative marks. When an examiner looks at two device marks and determines that they share a “similar overall impression”, they are making a structural judgement about composition, proportion, and arrangement, not counting matching pixels.

Validation against real decisions

The methodology has been validated against published tribunal and Board of Appeal decisions. This means the system’s similarity assessments have been compared against cases where human adjudicators examined pairs of marks and reached conclusions about visual similarity. The validation is against how tribunals actually decide, not against synthetic test sets or internal benchmarks.

This validation approach is important because trade mark similarity is ultimately a legal question answered by human decision-makers. A system that agrees with those decision-makers on cases they have already decided is more likely to produce useful assessments on new cases than one validated only against its own training data.

Explainability as a requirement

A similarity score alone is not useful in trade mark practice. When an adviser needs to explain to a client why two marks conflict, or when an opposition needs to be supported with evidence, the reasoning behind the assessment matters as much as the conclusion.

The geometric approach produces structured explanations. It can identify which elements of two marks are structurally equivalent, where they diverge, and how significant the similarities are relative to the overall composition. This means the analysis can be reviewed, challenged, and presented in proceedings, rather than accepted or rejected as an opaque number.

The published research behind this approach is available for examination. The methodology is not hidden behind a proprietary system or accessible only through a sales process. The science is open, the validation is documented, and the outputs are explainable. That transparency is not incidental to the approach. It is central to it.